A hybrid partitioned deep learning methodology for moving interface and fluid–structure interaction

In this work, we present a hybrid partitioned deep learning framework for the reduced-order modeling of moving interfaces and predicting fluid–structure interaction. Using discretized Navier–Stokes in arbitrary Lagrangian–Eulerian reference frame, generate full-order flow snapshots point cloud displacements as target physical data inference coupled dynamics. The operation methodology comes by combining two separate data-driven models fluid solid subdomains via learning-based (DL-ROMs). proposed multi-level comprises drivers unsteady displacements. At interface, force information is exchanged synchronously between subdomain solvers. first component our relies on proper orthogonal decomposition-based recurrent neural network (POD-RNN) DL-ROM procedure to infer with interface. This model utilizes POD basis modes reduce dimensionality evolve them time long short-term memory-based networks (LSTM-RNNs). second employs convolution-based autoencoder (CRAN) self-supervised nonlinear dynamics at static Eulerian probes. We introduce these probes spatially structured query nodes treat Lagrangian-to-Eulerian conflict together convenience training CRAN driver. To determine probes, construct novel snapshot-field transfer load recovery algorithm. They are chosen such way that components (i.e., POD-RNN CRAN) constrained interface recover bulk quantities. These DL-ROM-based rely LSTM-RNNs low-dimensional states. A popular prototypical interaction problem past freely oscillating cylinder considered assess efficacy different set reduced velocities lead vortex-induced vibrations. tracks description acceptable accuracy predicts wake over test range. aligns development digital twin engineering systems, especially those involving boundaries interactions.